A client holding a row lock can cause node restart (CopyFrag) to stall until the row lock is released. Check that a stalled CopyFrag does not lead to TC takeover stall by : - Restarting a node - Stalling CopyFrag due to a rowlock - Causing starting node failure - Checking that TC takeover completes (As transactions can be committed in different epochs) create table t1(id int primary key, val int) engine=ndb; select count(*) from t1; count(*) 10 begin; select * from t1 where id = 1 for update; id val 1 1 select tc_node_id from ndbinfo.cluster_operations; tc_node_id 1 select ndbinfo.cluster_transactions.state, ndbinfo.cluster_operations.operation_type, ndbinfo.cluster_operations.state, ndbinfo.cluster_locks.mode from ndbinfo.cluster_transactions, ndbinfo.cluster_operations, ndbinfo.cluster_locks where ndbinfo.cluster_transactions.transid = ndbinfo.cluster_locks.transid and ndbinfo.cluster_locks.transid = ndbinfo.cluster_operations.transid; state operation_type state mode Started READ-EX Prepared X Starting the nowait node Eventually it will be waiting for the lock above Kill the starting node This will result in TC takeover Sleeping 10 s for node failure to be detected and handling to start [and finish] Test succeeded, no GCP stall Release lock to allow data node start to succeed rollback; Wait for all data nodes to start Started ok drop table t1;